Computerized Bone Age Estimation Using Deep Learning Based Program: Evaluation of the Accuracy and Efficiency

Area of Science:

• Radiology
• Artificial Intelligence in Medicine
• Pediatric Endocrinology

Background:

• Accurate bone age assessment is crucial for evaluating growth and development in children.
• Traditional methods, like the Greulich-Pyle atlas, can be time-consuming and subjective.
• Advancements in artificial intelligence offer potential for automated and more efficient assessments.

Purpose of the Study:

• To evaluate the accuracy and efficiency of a novel deep learning-based automatic software system for bone age assessment.
• To validate the system's feasibility for clinical application in pediatric patients.
• To compare the performance of the automatic system against traditional methods and radiologist interpretation.

Main Methods:

• A deep learning model was developed based on the Greulich-Pyle method for automated bone age determination.
• Left-hand radiographs of 200 pediatric patients (ages 3-17) were analyzed.
• Bone age was assessed using: software-only (first-rank), computer-assisted (radiologists + software), and atlas-assisted (radiologists + Greulich-Pyle atlas).
• Reference bone age was established by consensus of two experienced radiologists.

Main Results:

• The automatic software system achieved a 69.5% concordance rate with reference bone age, showing a strong correlation (r = 0.992, p < 0.001).
• Computer-assisted assessment using the software improved concordance rates for both radiologists compared to atlas-assisted assessment.
• Reading times were significantly reduced by 18.0% and 40.0% for the two radiologists, respectively.

Conclusions:

• The automatic software system provides reliable and accurate bone age estimations.
• The system enhances efficiency by decreasing reading times.
• The technology shows promise for clinical integration without sacrificing diagnostic accuracy in bone age assessment.